From bc51c6eb25d6ac62b8e72c90d2d1fe4d567f495e Mon Sep 17 00:00:00 2001 From: hamadmarri Date: Wed, 28 Jul 2021 06:17:36 +0300 Subject: [PATCH] Experimental: Multi factors scores --- .../CacULE/experimental/cacule-5.13-mfs.patch | 1385 +++++++++++++++++ .../CacULE/experimental/rdb-5.13-mfs.patch | 918 +++++++++++ 2 files changed, 2303 insertions(+) create mode 100644 patches/CacULE/experimental/cacule-5.13-mfs.patch create mode 100644 patches/CacULE/experimental/rdb-5.13-mfs.patch diff --git a/patches/CacULE/experimental/cacule-5.13-mfs.patch b/patches/CacULE/experimental/cacule-5.13-mfs.patch new file mode 100644 index 000000000..49f41626c --- /dev/null +++ b/patches/CacULE/experimental/cacule-5.13-mfs.patch @@ -0,0 +1,1385 @@ +diff --git a/Documentation/admin-guide/sysctl/kernel.rst b/Documentation/admin-guide/sysctl/kernel.rst +index 68b21395a743..3f4b9c6911be 100644 +--- a/Documentation/admin-guide/sysctl/kernel.rst ++++ b/Documentation/admin-guide/sysctl/kernel.rst +@@ -1088,6 +1088,10 @@ Model available). If your platform happens to meet the + requirements for EAS but you do not want to use it, change + this value to 0. + ++sched_interactivity_factor (CacULE scheduler only) ++================================================== ++Sets the value *m* for interactivity score calculations. See ++Figure 1 in https://web.cs.ucdavis.edu/~roper/ecs150/ULE.pdf + + sched_schedstats + ================ +diff --git a/Documentation/scheduler/sched-CacULE.rst b/Documentation/scheduler/sched-CacULE.rst +new file mode 100644 +index 000000000000..82b0847c468a +--- /dev/null ++++ b/Documentation/scheduler/sched-CacULE.rst +@@ -0,0 +1,76 @@ ++====================================== ++The CacULE Scheduler by Hamad Al Marri. ++====================================== ++ ++1. Overview ++============= ++ ++The CacULE CPU scheduler is based on interactivity score mechanism. ++The interactivity score is inspired by the ULE scheduler (FreeBSD ++scheduler). ++ ++1.1 About CacULE Scheduler ++-------------------------- ++ ++ - Each CPU has its own runqueue. ++ ++ - NORMAL runqueue is a linked list of sched_entities (instead of RB-Tree). ++ ++ - RT and other runqueues are just the same as the CFS's. ++ ++ - Wake up tasks preempt currently running tasks if its interactivity score value ++ is higher. ++ ++ ++1.2. Complexity ++---------------- ++ ++The complexity of Enqueue and Dequeue a task is O(1). ++ ++The complexity of pick the next task is in O(n), where n is the number of tasks ++in a runqueue (each CPU has its own runqueue). ++ ++Note: O(n) sounds scary, but usually for a machine with 4 CPUS where it is used ++for desktop or mobile jobs, the maximum number of runnable tasks might not ++exceeds 10 (at the pick next run time) - the idle tasks are excluded since they ++are dequeued when sleeping and enqueued when they wake up. ++ ++ ++2. The CacULE Interactivity Score ++======================================================= ++ ++The interactivity score is inspired by the ULE scheduler (FreeBSD scheduler). ++For more information see: https://web.cs.ucdavis.edu/~roper/ecs150/ULE.pdf ++CacULE doesn't replace CFS with ULE, it only changes the CFS' pick next task ++mechanism to ULE's interactivity score mechanism for picking next task to run. ++ ++ ++2.3 sched_interactivity_factor ++================= ++Sets the value *m* for interactivity score calculations. See Figure 1 in ++https://web.cs.ucdavis.edu/~roper/ecs150/ULE.pdf ++The default value of in CacULE is 10 which means that the Maximum Interactive ++Score is 20 (since m = Maximum Interactive Score / 2). ++You can tune sched_interactivity_factor with sysctl command: ++ ++ sysctl kernel.sched_interactivity_factor=50 ++ ++This command changes the sched_interactivity_factor from 10 to 50. ++ ++ ++3. Scheduling policies ++======================= ++ ++CacULE some CFS, implements three scheduling policies: ++ ++ - SCHED_NORMAL (traditionally called SCHED_OTHER): The scheduling ++ policy that is used for regular tasks. ++ ++ - SCHED_BATCH: Does not preempt nearly as often as regular tasks ++ would, thereby allowing tasks to run longer and make better use of ++ caches but at the cost of interactivity. This is well suited for ++ batch jobs. ++ ++ - SCHED_IDLE: This is even weaker than nice 19, but its not a true ++ idle timer scheduler in order to avoid to get into priority ++ inversion problems which would deadlock the machine. +diff --git a/include/linux/sched.h b/include/linux/sched.h +index 32813c345115..0dc06f09715f 100644 +--- a/include/linux/sched.h ++++ b/include/linux/sched.h +@@ -458,10 +458,23 @@ struct sched_statistics { + #endif + }; + ++#ifdef CONFIG_CACULE_SCHED ++struct cacule_node { ++ struct cacule_node* next; ++ struct cacule_node* prev; ++ u64 cacule_start_time; ++ u64 last_run; ++ u64 vruntime; ++}; ++#endif ++ + struct sched_entity { + /* For load-balancing: */ + struct load_weight load; + struct rb_node run_node; ++#ifdef CONFIG_CACULE_SCHED ++ struct cacule_node cacule_node; ++#endif + struct list_head group_node; + unsigned int on_rq; + +diff --git a/include/linux/sched/sysctl.h b/include/linux/sched/sysctl.h +index db2c0f34aaaf..05329f07d2a4 100644 +--- a/include/linux/sched/sysctl.h ++++ b/include/linux/sched/sysctl.h +@@ -32,6 +32,15 @@ extern unsigned int sysctl_sched_latency; + extern unsigned int sysctl_sched_min_granularity; + extern unsigned int sysctl_sched_wakeup_granularity; + ++#ifdef CONFIG_CACULE_SCHED ++extern unsigned int interactivity_factor; ++extern unsigned int cacule_max_lifetime; ++extern unsigned int cache_factor; ++extern unsigned int cache_divisor; ++extern unsigned int starve_factor; ++extern unsigned int starve_divisor; ++#endif ++ + enum sched_tunable_scaling { + SCHED_TUNABLESCALING_NONE, + SCHED_TUNABLESCALING_LOG, +diff --git a/init/Kconfig b/init/Kconfig +index a61c92066c2e..089e3bfe5dbc 100644 +--- a/init/Kconfig ++++ b/init/Kconfig +@@ -834,6 +834,17 @@ config UCLAMP_BUCKETS_COUNT + + endmenu + ++config CACULE_SCHED ++ bool "CacULE CPU scheduler" ++ default y ++ help ++ The CacULE CPU scheduler is based on interactivity score mechanism. ++ The interactivity score is inspired by the ULE scheduler (FreeBSD ++ scheduler). ++ ++ If unsure, say Y here. ++ ++ + # + # For architectures that want to enable the support for NUMA-affine scheduler + # balancing logic: +@@ -1231,6 +1242,7 @@ config SCHED_AUTOGROUP + select CGROUPS + select CGROUP_SCHED + select FAIR_GROUP_SCHED ++ default y + help + This option optimizes the scheduler for common desktop workloads by + automatically creating and populating task groups. This separation +diff --git a/kernel/Kconfig.hz b/kernel/Kconfig.hz +index 38ef6d06888e..865f8dbddca8 100644 +--- a/kernel/Kconfig.hz ++++ b/kernel/Kconfig.hz +@@ -46,6 +46,9 @@ choice + 1000 Hz is the preferred choice for desktop systems and other + systems requiring fast interactive responses to events. + ++ config HZ_2000 ++ bool "2000 HZ" ++ + endchoice + + config HZ +@@ -54,6 +57,7 @@ config HZ + default 250 if HZ_250 + default 300 if HZ_300 + default 1000 if HZ_1000 ++ default 2000 if HZ_2000 + + config SCHED_HRTICK + def_bool HIGH_RES_TIMERS +diff --git a/kernel/sched/core.c b/kernel/sched/core.c +index e5858999b54d..bdedde199504 100644 +--- a/kernel/sched/core.c ++++ b/kernel/sched/core.c +@@ -3578,6 +3578,11 @@ static void __sched_fork(unsigned long clone_flags, struct task_struct *p) + p->se.prev_sum_exec_runtime = 0; + p->se.nr_migrations = 0; + p->se.vruntime = 0; ++ ++#ifdef CONFIG_CACULE_SCHED ++ p->se.cacule_node.vruntime = 0; ++#endif ++ + INIT_LIST_HEAD(&p->se.group_node); + + #ifdef CONFIG_FAIR_GROUP_SCHED +@@ -3863,6 +3868,10 @@ void wake_up_new_task(struct task_struct *p) + update_rq_clock(rq); + post_init_entity_util_avg(p); + ++#ifdef CONFIG_CACULE_SCHED ++ p->se.cacule_node.cacule_start_time = sched_clock(); ++#endif ++ + activate_task(rq, p, ENQUEUE_NOCLOCK); + trace_sched_wakeup_new(p); + check_preempt_curr(rq, p, WF_FORK); +@@ -4674,7 +4683,9 @@ static void sched_tick_remote(struct work_struct *work) + struct rq *rq = cpu_rq(cpu); + struct task_struct *curr; + struct rq_flags rf; ++#if !defined(CONFIG_CACULE_SCHED) + u64 delta; ++#endif + int os; + + /* +@@ -4694,6 +4705,7 @@ static void sched_tick_remote(struct work_struct *work) + + update_rq_clock(rq); + ++#if !defined(CONFIG_CACULE_SCHED) + if (!is_idle_task(curr)) { + /* + * Make sure the next tick runs within a reasonable +@@ -4702,6 +4714,8 @@ static void sched_tick_remote(struct work_struct *work) + delta = rq_clock_task(rq) - curr->se.exec_start; + WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 3); + } ++#endif ++ + curr->sched_class->task_tick(rq, curr, 0); + + calc_load_nohz_remote(rq); +@@ -8115,6 +8129,10 @@ void __init sched_init(void) + BUG_ON(&dl_sched_class + 1 != &stop_sched_class); + #endif + ++#ifdef CONFIG_CACULE_SCHED ++ printk(KERN_INFO "CacULE CPU scheduler v5.13-r2 by Hamad Al Marri."); ++#endif ++ + wait_bit_init(); + + #ifdef CONFIG_FAIR_GROUP_SCHED +diff --git a/kernel/sched/debug.c b/kernel/sched/debug.c +index c5aacbd492a1..adb021b7da8a 100644 +--- a/kernel/sched/debug.c ++++ b/kernel/sched/debug.c +@@ -560,8 +560,11 @@ static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu) + + void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq) + { +- s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1, +- spread, rq0_min_vruntime, spread0; ++ s64 MIN_vruntime = -1, max_vruntime = -1, ++#if !defined(CONFIG_CACULE_SCHED) ++ min_vruntime, rq0_min_vruntime, spread0, ++#endif ++ spread; + struct rq *rq = cpu_rq(cpu); + struct sched_entity *last; + unsigned long flags; +@@ -582,21 +585,27 @@ void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq) + last = __pick_last_entity(cfs_rq); + if (last) + max_vruntime = last->vruntime; ++#if !defined(CONFIG_CACULE_SCHED) + min_vruntime = cfs_rq->min_vruntime; + rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime; ++#endif + raw_spin_unlock_irqrestore(&rq->lock, flags); + SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime", + SPLIT_NS(MIN_vruntime)); ++#if !defined(CONFIG_CACULE_SCHED) + SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime", + SPLIT_NS(min_vruntime)); ++#endif + SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime", + SPLIT_NS(max_vruntime)); + spread = max_vruntime - MIN_vruntime; + SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread", + SPLIT_NS(spread)); ++#if !defined(CONFIG_CACULE_SCHED) + spread0 = min_vruntime - rq0_min_vruntime; + SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0", + SPLIT_NS(spread0)); ++#endif + SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over", + cfs_rq->nr_spread_over); + SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running); +diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c +index 7dd0d859d95b..c64337fd121d 100644 +--- a/kernel/sched/fair.c ++++ b/kernel/sched/fair.c +@@ -19,9 +19,24 @@ + * + * Adaptive scheduling granularity, math enhancements by Peter Zijlstra + * Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra ++ * ++ * CacULE enhancements CPU cache and scheduler based on ++ * Interactivity Score. ++ * (C) 2020 Hamad Al Marri + */ + #include "sched.h" + ++#ifdef CONFIG_CACULE_SCHED ++unsigned int __read_mostly cacule_max_lifetime = 22000; // in ms ++unsigned int __read_mostly interactivity_factor = 32768; ++ ++unsigned int __read_mostly cache_factor = 13107; ++unsigned int __read_mostly cache_divisor = 1000000; // 1ms ++ ++unsigned int __read_mostly starve_factor = 19660; ++unsigned int __read_mostly starve_divisor = 3000000; // 3ms ++#endif ++ + /* + * Targeted preemption latency for CPU-bound tasks: + * +@@ -82,7 +97,11 @@ unsigned int sysctl_sched_child_runs_first __read_mostly; + unsigned int sysctl_sched_wakeup_granularity = 1000000UL; + static unsigned int normalized_sysctl_sched_wakeup_granularity = 1000000UL; + ++#ifdef CONFIG_CACULE_SCHED ++const_debug unsigned int sysctl_sched_migration_cost = 200000UL; ++#else + const_debug unsigned int sysctl_sched_migration_cost = 500000UL; ++#endif + + int sched_thermal_decay_shift; + static int __init setup_sched_thermal_decay_shift(char *str) +@@ -263,6 +282,14 @@ static u64 __calc_delta(u64 delta_exec, unsigned long weight, struct load_weight + + const struct sched_class fair_sched_class; + ++ ++#ifdef CONFIG_CACULE_SCHED ++static inline struct sched_entity *se_of(struct cacule_node *cn) ++{ ++ return container_of(cn, struct sched_entity, cacule_node); ++} ++#endif ++ + /************************************************************** + * CFS operations on generic schedulable entities: + */ +@@ -522,7 +549,7 @@ void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec); + /************************************************************** + * Scheduling class tree data structure manipulation methods: + */ +- ++#if !defined(CONFIG_CACULE_SCHED) + static inline u64 max_vruntime(u64 max_vruntime, u64 vruntime) + { + s64 delta = (s64)(vruntime - max_vruntime); +@@ -585,7 +612,260 @@ static inline bool __entity_less(struct rb_node *a, const struct rb_node *b) + { + return entity_before(__node_2_se(a), __node_2_se(b)); + } ++#endif /* CONFIG_CACULE_SCHED */ ++ ++#ifdef CONFIG_CACULE_SCHED ++static unsigned int ++calc_interactivity(u64 now, struct cacule_node *se) ++{ ++ u64 l_se, vr_se, sleep_se = 1ULL, u64_factor_m, _2m; ++ unsigned int score_se; ++ ++ /* ++ * in case of vruntime==0, logical OR with 1 would ++ * make sure that the least sig. bit is 1 ++ */ ++ l_se = now - se->cacule_start_time; ++ vr_se = se->vruntime | 1; ++ u64_factor_m = interactivity_factor; ++ _2m = u64_factor_m << 1; ++ ++ /* safety check */ ++ if (likely(l_se > vr_se)) ++ sleep_se = (l_se - vr_se) | 1; ++ ++ if (sleep_se >= vr_se) ++ score_se = u64_factor_m / (sleep_se / vr_se); ++ else ++ score_se = _2m - (u64_factor_m / (vr_se / sleep_se)); ++ ++ return score_se; ++} ++ ++static unsigned int ++calc_cache_score(u64 now, struct cacule_node *cn) ++{ ++ struct sched_entity *se = se_of(cn); ++ struct cfs_rq *cfs_rq = cfs_rq_of(se); ++ u64 c_div = cache_divisor; ++ u64 cache_period = 1ULL; ++ u64 u64_factor_m = cache_factor; ++ u64 _2m = u64_factor_m << 1; ++ unsigned int score; ++ ++ if (!cache_factor) ++ return 0; ++ ++ if (se == cfs_rq->curr) ++ return 0; ++ ++ cache_period = (now - se->exec_start) | 1; ++ ++ if (c_div >= cache_period) ++ score = u64_factor_m / (c_div / cache_period); ++ else ++ score = _2m - (u64_factor_m / (cache_period / c_div)); ++ ++ return score; ++} ++ ++static unsigned int ++calc_starve_score(u64 now, struct cacule_node *cn) ++{ ++ struct sched_entity *se = se_of(cn); ++ struct cfs_rq *cfs_rq = cfs_rq_of(se); ++ u64 s_div = starve_divisor; ++ u64 starving = 1ULL; ++ u64 u64_factor_m = starve_factor; ++ u64 _2m = u64_factor_m << 1; ++ unsigned int score; ++ ++ if (!starve_factor) ++ return 0; ++ ++ if (se == cfs_rq->curr) ++ return _2m; ++ ++ starving = (now - cn->last_run) | 1; ++ ++ if (s_div >= starving) ++ score = _2m - (u64_factor_m / (s_div / starving)); ++ else ++ score = u64_factor_m / (starving / s_div); ++ ++ return score; ++} ++ ++static inline int cn_has_idle_policy(struct cacule_node *cn) ++{ ++ struct sched_entity *se = se_of(cn); ++ ++ if (!entity_is_task(se)) ++ return false; ++ ++ return task_has_idle_policy(task_of(se)); ++} ++ ++static inline int ++entity_before_cached(u64 now, unsigned int score_curr, struct cacule_node *se) ++{ ++ unsigned int score_se; ++ int diff; ++ ++ /* ++ * if se has idle class, then no need to ++ * calculate, since we are sure that score_curr ++ * is a score for non idle class task ++ */ ++ if (cn_has_idle_policy(se)) ++ return -1; ++ ++ score_se = calc_interactivity(now, se); ++ diff = score_se - score_curr; ++ ++ if (diff <= 0) ++ return 1; ++ ++ return -1; ++} + ++/* ++ * Does se have lower interactivity score value (i.e. interactive) than curr? If yes, return 1, ++ * otherwise return -1 ++ * se is before curr if se has lower interactivity score value ++ * the lower score, the more interactive ++ */ ++static inline int ++entity_before(u64 now, struct cacule_node *curr, struct cacule_node *se) ++{ ++ unsigned int score_curr, score_se; ++ int diff; ++ int is_curr_idle = cn_has_idle_policy(curr); ++ int is_se_idle = cn_has_idle_policy(se); ++ ++ /* if curr is normal but se is idle class, then no */ ++ if (!is_curr_idle && is_se_idle) ++ return -1; ++ ++ /* if curr is idle class and se is normal, then yes */ ++ if (is_curr_idle && !is_se_idle) ++ return 1; ++ ++ score_curr = calc_interactivity(now, curr); ++ score_curr += calc_cache_score(now, curr); ++ score_curr += calc_starve_score(now, curr); ++ ++ score_se = calc_interactivity(now, se); ++ score_se += calc_cache_score(now, se); ++ score_se += calc_starve_score(now, se); ++ ++ diff = score_se - score_curr; ++ ++ if (diff < 0) ++ return 1; ++ ++ return -1; ++} ++ ++/* ++ * Enqueue an entity ++ */ ++static void __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *_se) ++{ ++ struct cacule_node *se = &(_se->cacule_node); ++ struct cacule_node *iter, *next = NULL; ++ u64 now = sched_clock(); ++ unsigned int score_se = calc_interactivity(now, se); ++ int is_idle_task = cn_has_idle_policy(se); ++ ++ se->next = NULL; ++ se->prev = NULL; ++ ++ if (likely(cfs_rq->head)) { ++ ++ // start from tail ++ iter = cfs_rq->tail; ++ ++ /* ++ * if this task has idle class, then ++ * push it to the tail right away ++ */ ++ if (is_idle_task) ++ goto to_tail; ++ ++ /* here we know that this task isn't idle clas */ ++ ++ // does se have higher IS than iter? ++ while (iter && entity_before_cached(now, score_se, iter) == -1) { ++ next = iter; ++ iter = iter->prev; ++ } ++ ++ // se in tail position ++ if (iter == cfs_rq->tail) { ++to_tail: ++ cfs_rq->tail->next = se; ++ se->prev = cfs_rq->tail; ++ ++ cfs_rq->tail = se; ++ } ++ // else if not head no tail, insert se after iter ++ else if (iter) { ++ se->next = next; ++ se->prev = iter; ++ ++ iter->next = se; ++ next->prev = se; ++ } ++ // insert se at head ++ else { ++ se->next = cfs_rq->head; ++ cfs_rq->head->prev = se; ++ ++ // lastly reset the head ++ cfs_rq->head = se; ++ } ++ } else { ++ // if empty rq ++ cfs_rq->head = se; ++ cfs_rq->tail = se; ++ } ++} ++ ++static void __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *_se) ++{ ++ struct cacule_node *se = &(_se->cacule_node); ++ ++ // if only one se in rq ++ if (cfs_rq->head == cfs_rq->tail) { ++ cfs_rq->head = NULL; ++ cfs_rq->tail = NULL; ++ ++ } else if (se == cfs_rq->head) { ++ // if it is the head ++ cfs_rq->head = cfs_rq->head->next; ++ cfs_rq->head->prev = NULL; ++ } else if (se == cfs_rq->tail) { ++ // if it is the tail ++ cfs_rq->tail = cfs_rq->tail->prev; ++ cfs_rq->tail->next = NULL; ++ } else { ++ // if in the middle ++ struct cacule_node *prev = se->prev; ++ struct cacule_node *next = se->next; ++ ++ prev->next = next; ++ ++ if (next) ++ next->prev = prev; ++ } ++} ++ ++struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq) ++{ ++ return se_of(cfs_rq->head); ++} ++#else + /* + * Enqueue an entity into the rb-tree: + */ +@@ -618,16 +898,24 @@ static struct sched_entity *__pick_next_entity(struct sched_entity *se) + + return __node_2_se(next); + } ++#endif /* CONFIG_CACULE_SCHED */ + + #ifdef CONFIG_SCHED_DEBUG + struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq) + { ++#ifdef CONFIG_CACULE_SCHED ++ if (!cfs_rq->tail) ++ return NULL; ++ ++ return se_of(cfs_rq->tail); ++#else + struct rb_node *last = rb_last(&cfs_rq->tasks_timeline.rb_root); + + if (!last) + return NULL; + + return __node_2_se(last); ++#endif /* CONFIG_CACULE_SCHED */ + } + + /************************************************************** +@@ -717,6 +1005,7 @@ static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se) + return slice; + } + ++#if !defined(CONFIG_CACULE_SCHED) + /* + * We calculate the vruntime slice of a to-be-inserted task. + * +@@ -726,6 +1015,7 @@ static u64 sched_vslice(struct cfs_rq *cfs_rq, struct sched_entity *se) + { + return calc_delta_fair(sched_slice(cfs_rq, se), se); + } ++#endif /* CONFIG_CACULE_SCHED */ + + #include "pelt.h" + #ifdef CONFIG_SMP +@@ -833,14 +1123,51 @@ static void update_tg_load_avg(struct cfs_rq *cfs_rq) + } + #endif /* CONFIG_SMP */ + ++#ifdef CONFIG_CACULE_SCHED ++static void normalize_lifetime(u64 now, struct sched_entity *se) ++{ ++ struct cacule_node *cn = &se->cacule_node; ++ u64 max_life_ns, life_time; ++ s64 diff; ++ ++ /* ++ * left shift 20 bits is approximately = * 1000000 ++ * we don't need the precision of life time ++ * Ex. for 30s, with left shift (20bits) == 31.457s ++ */ ++ max_life_ns = ((u64) cacule_max_lifetime) << 20; ++ life_time = now - cn->cacule_start_time; ++ diff = life_time - max_life_ns; ++ ++ if (diff > 0) { ++ // multiply life_time by 1024 for more precision ++ u64 old_hrrn_x = (life_time << 7) / ((cn->vruntime >> 3) | 1); ++ ++ // reset life to half max_life (i.e ~15s) ++ cn->cacule_start_time = now - (max_life_ns >> 1); ++ ++ // avoid division by zero ++ if (old_hrrn_x == 0) old_hrrn_x = 1; ++ ++ // reset vruntime based on old hrrn ratio ++ cn->vruntime = (max_life_ns << 9) / old_hrrn_x; ++ } ++} ++#endif /* CONFIG_CACULE_SCHED */ ++ + /* + * Update the current task's runtime statistics. + */ + static void update_curr(struct cfs_rq *cfs_rq) + { + struct sched_entity *curr = cfs_rq->curr; ++#ifdef CONFIG_CACULE_SCHED ++ u64 now = sched_clock(); ++ u64 delta_exec, delta_fair; ++#else + u64 now = rq_clock_task(rq_of(cfs_rq)); + u64 delta_exec; ++#endif + + if (unlikely(!curr)) + return; +@@ -857,8 +1184,16 @@ static void update_curr(struct cfs_rq *cfs_rq) + curr->sum_exec_runtime += delta_exec; + schedstat_add(cfs_rq->exec_clock, delta_exec); + ++#ifdef CONFIG_CACULE_SCHED ++ curr->cacule_node.last_run = now; ++ delta_fair = calc_delta_fair(delta_exec, curr); ++ curr->vruntime += delta_fair; ++ curr->cacule_node.vruntime += delta_fair; ++ normalize_lifetime(now, curr); ++#else + curr->vruntime += calc_delta_fair(delta_exec, curr); + update_min_vruntime(cfs_rq); ++#endif + + if (entity_is_task(curr)) { + struct task_struct *curtask = task_of(curr); +@@ -1026,7 +1361,6 @@ update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) + static inline void + update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) + { +- + if (!schedstat_enabled()) + return; + +@@ -1058,7 +1392,12 @@ update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se) + /* + * We are starting a new run period: + */ ++#ifdef CONFIG_CACULE_SCHED ++ se->exec_start = sched_clock(); ++ se->cacule_node.last_run = sched_clock(); ++#else + se->exec_start = rq_clock_task(rq_of(cfs_rq)); ++#endif + } + + /************************************************** +@@ -4178,7 +4517,7 @@ static inline void update_misfit_status(struct task_struct *p, struct rq *rq) {} + + static void check_spread(struct cfs_rq *cfs_rq, struct sched_entity *se) + { +-#ifdef CONFIG_SCHED_DEBUG ++#if defined(CONFIG_SCHED_DEBUG) && !defined(CONFIG_CACULE_SCHED) + s64 d = se->vruntime - cfs_rq->min_vruntime; + + if (d < 0) +@@ -4189,6 +4528,7 @@ static void check_spread(struct cfs_rq *cfs_rq, struct sched_entity *se) + #endif + } + ++#if !defined(CONFIG_CACULE_SCHED) + static void + place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial) + { +@@ -4220,6 +4560,7 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial) + /* ensure we never gain time by being placed backwards. */ + se->vruntime = max_vruntime(se->vruntime, vruntime); + } ++#endif /* CONFIG_CACULE_SCHED */ + + static void check_enqueue_throttle(struct cfs_rq *cfs_rq); + +@@ -4278,18 +4619,23 @@ static inline bool cfs_bandwidth_used(void); + static void + enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) + { ++#if !defined(CONFIG_CACULE_SCHED) + bool renorm = !(flags & ENQUEUE_WAKEUP) || (flags & ENQUEUE_MIGRATED); ++#endif + bool curr = cfs_rq->curr == se; + ++#if !defined(CONFIG_CACULE_SCHED) + /* + * If we're the current task, we must renormalise before calling + * update_curr(). + */ + if (renorm && curr) + se->vruntime += cfs_rq->min_vruntime; ++#endif + + update_curr(cfs_rq); + ++#if !defined(CONFIG_CACULE_SCHED) + /* + * Otherwise, renormalise after, such that we're placed at the current + * moment in time, instead of some random moment in the past. Being +@@ -4298,6 +4644,7 @@ enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) + */ + if (renorm && !curr) + se->vruntime += cfs_rq->min_vruntime; ++#endif + + /* + * When enqueuing a sched_entity, we must: +@@ -4312,8 +4659,10 @@ enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) + update_cfs_group(se); + account_entity_enqueue(cfs_rq, se); + ++#if !defined(CONFIG_CACULE_SCHED) + if (flags & ENQUEUE_WAKEUP) + place_entity(cfs_rq, se, 0); ++#endif + + check_schedstat_required(); + update_stats_enqueue(cfs_rq, se, flags); +@@ -4334,6 +4683,7 @@ enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) + check_enqueue_throttle(cfs_rq); + } + ++#if !defined(CONFIG_CACULE_SCHED) + static void __clear_buddies_last(struct sched_entity *se) + { + for_each_sched_entity(se) { +@@ -4378,6 +4728,7 @@ static void clear_buddies(struct cfs_rq *cfs_rq, struct sched_entity *se) + if (cfs_rq->skip == se) + __clear_buddies_skip(se); + } ++#endif /* !CONFIG_CACULE_SCHED */ + + static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq); + +@@ -4402,13 +4753,16 @@ dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) + + update_stats_dequeue(cfs_rq, se, flags); + ++#if !defined(CONFIG_CACULE_SCHED) + clear_buddies(cfs_rq, se); ++#endif + + if (se != cfs_rq->curr) + __dequeue_entity(cfs_rq, se); + se->on_rq = 0; + account_entity_dequeue(cfs_rq, se); + ++#if !defined(CONFIG_CACULE_SCHED) + /* + * Normalize after update_curr(); which will also have moved + * min_vruntime if @se is the one holding it back. But before doing +@@ -4417,12 +4771,14 @@ dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) + */ + if (!(flags & DEQUEUE_SLEEP)) + se->vruntime -= cfs_rq->min_vruntime; ++#endif + + /* return excess runtime on last dequeue */ + return_cfs_rq_runtime(cfs_rq); + + update_cfs_group(se); + ++#if !defined(CONFIG_CACULE_SCHED) + /* + * Now advance min_vruntime if @se was the entity holding it back, + * except when: DEQUEUE_SAVE && !DEQUEUE_MOVE, in this case we'll be +@@ -4431,8 +4787,23 @@ dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) + */ + if ((flags & (DEQUEUE_SAVE | DEQUEUE_MOVE)) != DEQUEUE_SAVE) + update_min_vruntime(cfs_rq); ++#endif + } + ++#ifdef CONFIG_CACULE_SCHED ++static struct sched_entity * ++pick_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *curr); ++ ++/* ++ * Preempt the current task with a newly woken task if needed: ++ */ ++static void ++check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) ++{ ++ if (pick_next_entity(cfs_rq, curr) != curr) ++ resched_curr(rq_of(cfs_rq)); ++} ++#else + /* + * Preempt the current task with a newly woken task if needed: + */ +@@ -4472,6 +4843,7 @@ check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) + if (delta > ideal_runtime) + resched_curr(rq_of(cfs_rq)); + } ++#endif /* CONFIG_CACULE_SCHED */ + + static void + set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) +@@ -4506,6 +4878,31 @@ set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) + se->prev_sum_exec_runtime = se->sum_exec_runtime; + } + ++#ifdef CONFIG_CACULE_SCHED ++static struct sched_entity * ++pick_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *curr) ++{ ++ struct cacule_node *se = cfs_rq->head; ++ struct cacule_node *next; ++ u64 now = sched_clock(); ++ ++ if (unlikely(!se)) ++ return curr; ++ ++ next = se->next; ++ while (next) { ++ if (entity_before(now, se, next) == 1) ++ se = next; ++ ++ next = next->next; ++ } ++ ++ if (curr && entity_before(now, se, &curr->cacule_node) == 1) ++ return curr; ++ ++ return se_of(se); ++} ++#else + static int + wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se); + +@@ -4566,6 +4963,7 @@ pick_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *curr) + + return se; + } ++#endif /* CONFIG_CACULE_SCHED */ + + static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq); + +@@ -5668,7 +6066,9 @@ enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags) + hrtick_update(rq); + } + ++#if !defined(CONFIG_CACULE_SCHED) + static void set_next_buddy(struct sched_entity *se); ++#endif + + /* + * The dequeue_task method is called before nr_running is +@@ -5700,12 +6100,14 @@ static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags) + if (cfs_rq->load.weight) { + /* Avoid re-evaluating load for this entity: */ + se = parent_entity(se); ++#if !defined(CONFIG_CACULE_SCHED) + /* + * Bias pick_next to pick a task from this cfs_rq, as + * p is sleeping when it is within its sched_slice. + */ + if (task_sleep && se && !throttled_hierarchy(cfs_rq)) + set_next_buddy(se); ++#endif + break; + } + flags |= DEQUEUE_SLEEP; +@@ -5821,6 +6223,7 @@ static unsigned long capacity_of(int cpu) + return cpu_rq(cpu)->cpu_capacity; + } + ++#if !defined(CONFIG_CACULE_SCHED) + static void record_wakee(struct task_struct *p) + { + /* +@@ -5867,6 +6270,7 @@ static int wake_wide(struct task_struct *p) + return 0; + return 1; + } ++#endif /* CONFIG_CACULE_SCHED */ + + /* + * The purpose of wake_affine() is to quickly determine on which CPU we can run +@@ -6569,6 +6973,7 @@ static unsigned long cpu_util_without(int cpu, struct task_struct *p) + return min_t(unsigned long, util, capacity_orig_of(cpu)); + } + ++#if !defined(CONFIG_CACULE_SCHED) + /* + * Predicts what cpu_util(@cpu) would return if @p was migrated (and enqueued) + * to @dst_cpu. +@@ -6823,6 +7228,7 @@ static int find_energy_efficient_cpu(struct task_struct *p, int prev_cpu) + + return -1; + } ++#endif /* CONFIG_CACULE_SCHED */ + + /* + * select_task_rq_fair: Select target runqueue for the waking task in domains +@@ -6847,6 +7253,7 @@ select_task_rq_fair(struct task_struct *p, int prev_cpu, int wake_flags) + /* SD_flags and WF_flags share the first nibble */ + int sd_flag = wake_flags & 0xF; + ++#if !defined(CONFIG_CACULE_SCHED) + if (wake_flags & WF_TTWU) { + record_wakee(p); + +@@ -6859,6 +7266,7 @@ select_task_rq_fair(struct task_struct *p, int prev_cpu, int wake_flags) + + want_affine = !wake_wide(p) && cpumask_test_cpu(cpu, p->cpus_ptr); + } ++#endif /* CONFIG_CACULE_SCHED */ + + rcu_read_lock(); + for_each_domain(cpu, tmp) { +@@ -6905,6 +7313,7 @@ static void detach_entity_cfs_rq(struct sched_entity *se); + */ + static void migrate_task_rq_fair(struct task_struct *p, int new_cpu) + { ++#if !defined(CONFIG_CACULE_SCHED) + /* + * As blocked tasks retain absolute vruntime the migration needs to + * deal with this by subtracting the old and adding the new +@@ -6930,6 +7339,7 @@ static void migrate_task_rq_fair(struct task_struct *p, int new_cpu) + + se->vruntime -= min_vruntime; + } ++#endif /* CONFIG_CACULE_SCHED */ + + if (p->on_rq == TASK_ON_RQ_MIGRATING) { + /* +@@ -6975,6 +7385,7 @@ balance_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) + } + #endif /* CONFIG_SMP */ + ++#if !defined(CONFIG_CACULE_SCHED) + static unsigned long wakeup_gran(struct sched_entity *se) + { + unsigned long gran = sysctl_sched_wakeup_granularity; +@@ -7053,6 +7464,7 @@ static void set_skip_buddy(struct sched_entity *se) + for_each_sched_entity(se) + cfs_rq_of(se)->skip = se; + } ++#endif /* CONFIG_CACULE_SCHED */ + + /* + * Preempt the current task with a newly woken task if needed: +@@ -7061,9 +7473,12 @@ static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_ + { + struct task_struct *curr = rq->curr; + struct sched_entity *se = &curr->se, *pse = &p->se; ++ ++#if !defined(CONFIG_CACULE_SCHED) + struct cfs_rq *cfs_rq = task_cfs_rq(curr); + int scale = cfs_rq->nr_running >= sched_nr_latency; + int next_buddy_marked = 0; ++#endif /* CONFIG_CACULE_SCHED */ + + if (unlikely(se == pse)) + return; +@@ -7077,10 +7492,12 @@ static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_ + if (unlikely(throttled_hierarchy(cfs_rq_of(pse)))) + return; + ++#if !defined(CONFIG_CACULE_SCHED) + if (sched_feat(NEXT_BUDDY) && scale && !(wake_flags & WF_FORK)) { + set_next_buddy(pse); + next_buddy_marked = 1; + } ++#endif /* CONFIG_CACULE_SCHED */ + + /* + * We can come here with TIF_NEED_RESCHED already set from new task +@@ -7110,6 +7527,11 @@ static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_ + find_matching_se(&se, &pse); + update_curr(cfs_rq_of(se)); + BUG_ON(!pse); ++ ++#ifdef CONFIG_CACULE_SCHED ++ if (entity_before(sched_clock(), &se->cacule_node, &pse->cacule_node) == 1) ++ goto preempt; ++#else + if (wakeup_preempt_entity(se, pse) == 1) { + /* + * Bias pick_next to pick the sched entity that is +@@ -7119,11 +7541,14 @@ static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_ + set_next_buddy(pse); + goto preempt; + } ++#endif /* CONFIG_CACULE_SCHED */ + + return; + + preempt: + resched_curr(rq); ++ ++#if !defined(CONFIG_CACULE_SCHED) + /* + * Only set the backward buddy when the current task is still + * on the rq. This can happen when a wakeup gets interleaved +@@ -7138,6 +7563,7 @@ static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_ + + if (sched_feat(LAST_BUDDY) && scale && entity_is_task(se)) + set_last_buddy(se); ++#endif /* CONFIG_CACULE_SCHED */ + } + + struct task_struct * +@@ -7312,7 +7738,10 @@ static void yield_task_fair(struct rq *rq) + { + struct task_struct *curr = rq->curr; + struct cfs_rq *cfs_rq = task_cfs_rq(curr); ++ ++#if !defined(CONFIG_CACULE_SCHED) + struct sched_entity *se = &curr->se; ++#endif + + /* + * Are we the only task in the tree? +@@ -7320,7 +7749,9 @@ static void yield_task_fair(struct rq *rq) + if (unlikely(rq->nr_running == 1)) + return; + ++#if !defined(CONFIG_CACULE_SCHED) + clear_buddies(cfs_rq, se); ++#endif + + if (curr->policy != SCHED_BATCH) { + update_rq_clock(rq); +@@ -7336,7 +7767,9 @@ static void yield_task_fair(struct rq *rq) + rq_clock_skip_update(rq); + } + ++#if !defined(CONFIG_CACULE_SCHED) + set_skip_buddy(se); ++#endif + } + + static bool yield_to_task_fair(struct rq *rq, struct task_struct *p) +@@ -7347,8 +7780,10 @@ static bool yield_to_task_fair(struct rq *rq, struct task_struct *p) + if (!se->on_rq || throttled_hierarchy(cfs_rq_of(se))) + return false; + ++#if !defined(CONFIG_CACULE_SCHED) + /* Tell the scheduler that we'd really like pse to run next. */ + set_next_buddy(se); ++#endif + + yield_task_fair(rq); + +@@ -7575,6 +8010,7 @@ static int task_hot(struct task_struct *p, struct lb_env *env) + if (env->sd->flags & SD_SHARE_CPUCAPACITY) + return 0; + ++#if !defined(CONFIG_CACULE_SCHED) + /* + * Buddy candidates are cache hot: + */ +@@ -7582,6 +8018,7 @@ static int task_hot(struct task_struct *p, struct lb_env *env) + (&p->se == cfs_rq_of(&p->se)->next || + &p->se == cfs_rq_of(&p->se)->last)) + return 1; ++#endif + + if (sysctl_sched_migration_cost == -1) + return 1; +@@ -10592,9 +11029,11 @@ static void nohz_newidle_balance(struct rq *this_rq) + if (!housekeeping_cpu(this_cpu, HK_FLAG_SCHED)) + return; + ++#if !defined(CONFIG_CACULE_SCHED) + /* Will wake up very soon. No time for doing anything else*/ + if (this_rq->avg_idle < sysctl_sched_migration_cost) + return; ++#endif + + /* Don't need to update blocked load of idle CPUs*/ + if (!READ_ONCE(nohz.has_blocked) || +@@ -10657,7 +11096,10 @@ static int newidle_balance(struct rq *this_rq, struct rq_flags *rf) + */ + rq_unpin_lock(this_rq, rf); + +- if (this_rq->avg_idle < sysctl_sched_migration_cost || ++ if ( ++#if !defined(CONFIG_CACULE_SCHED) ++ this_rq->avg_idle < sysctl_sched_migration_cost || ++#endif + !READ_ONCE(this_rq->rd->overload)) { + + rcu_read_lock(); +@@ -10825,11 +11267,28 @@ static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued) + update_overutilized_status(task_rq(curr)); + } + ++#ifdef CONFIG_CACULE_SCHED + /* + * called on fork with the child task as argument from the parent's context + * - child not yet on the tasklist + * - preemption disabled + */ ++ static void task_fork_fair(struct task_struct *p) ++{ ++ struct cfs_rq *cfs_rq; ++ struct rq *rq = this_rq(); ++ struct rq_flags rf; ++ ++ rq_lock(rq, &rf); ++ update_rq_clock(rq); ++ ++ cfs_rq = task_cfs_rq(current); ++ if (cfs_rq->curr) ++ update_curr(cfs_rq); ++ ++ rq_unlock(rq, &rf); ++} ++#else + static void task_fork_fair(struct task_struct *p) + { + struct cfs_rq *cfs_rq; +@@ -10860,6 +11319,7 @@ static void task_fork_fair(struct task_struct *p) + se->vruntime -= cfs_rq->min_vruntime; + rq_unlock(rq, &rf); + } ++#endif /* CONFIG_CACULE_SCHED */ + + /* + * Priority of the task has changed. Check to see if we preempt +@@ -10978,6 +11438,8 @@ static void attach_entity_cfs_rq(struct sched_entity *se) + static void detach_task_cfs_rq(struct task_struct *p) + { + struct sched_entity *se = &p->se; ++ ++#if !defined(CONFIG_CACULE_SCHED) + struct cfs_rq *cfs_rq = cfs_rq_of(se); + + if (!vruntime_normalized(p)) { +@@ -10988,6 +11450,7 @@ static void detach_task_cfs_rq(struct task_struct *p) + place_entity(cfs_rq, se, 0); + se->vruntime -= cfs_rq->min_vruntime; + } ++#endif + + detach_entity_cfs_rq(se); + } +@@ -10995,12 +11458,17 @@ static void detach_task_cfs_rq(struct task_struct *p) + static void attach_task_cfs_rq(struct task_struct *p) + { + struct sched_entity *se = &p->se; ++ ++#if !defined(CONFIG_CACULE_SCHED) + struct cfs_rq *cfs_rq = cfs_rq_of(se); ++#endif + + attach_entity_cfs_rq(se); + ++#if !defined(CONFIG_CACULE_SCHED) + if (!vruntime_normalized(p)) + se->vruntime += cfs_rq->min_vruntime; ++#endif + } + + static void switched_from_fair(struct rq *rq, struct task_struct *p) +@@ -11056,13 +11524,22 @@ static void set_next_task_fair(struct rq *rq, struct task_struct *p, bool first) + void init_cfs_rq(struct cfs_rq *cfs_rq) + { + cfs_rq->tasks_timeline = RB_ROOT_CACHED; ++ ++#if !defined(CONFIG_CACULE_SCHED) + cfs_rq->min_vruntime = (u64)(-(1LL << 20)); + #ifndef CONFIG_64BIT + cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime; + #endif ++#endif /* CONFIG_CACULE_SCHED */ ++ + #ifdef CONFIG_SMP + raw_spin_lock_init(&cfs_rq->removed.lock); + #endif ++ ++#ifdef CONFIG_CACULE_SCHED ++ cfs_rq->head = NULL; ++ cfs_rq->tail = NULL; ++#endif + } + + #ifdef CONFIG_FAIR_GROUP_SCHED +diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h +index 35f7efed75c4..09a8290fc883 100644 +--- a/kernel/sched/sched.h ++++ b/kernel/sched/sched.h +@@ -525,10 +525,13 @@ struct cfs_rq { + unsigned int idle_h_nr_running; /* SCHED_IDLE */ + + u64 exec_clock; ++ ++#if !defined(CONFIG_CACULE_SCHED) + u64 min_vruntime; + #ifndef CONFIG_64BIT + u64 min_vruntime_copy; + #endif ++#endif /* CONFIG_CACULE_SCHED */ + + struct rb_root_cached tasks_timeline; + +@@ -537,9 +540,15 @@ struct cfs_rq { + * It is set to NULL otherwise (i.e when none are currently running). + */ + struct sched_entity *curr; ++#ifdef CONFIG_CACULE_SCHED ++ struct cacule_node *head; ++ struct cacule_node *tail; ++ ++#else + struct sched_entity *next; + struct sched_entity *last; + struct sched_entity *skip; ++#endif // CONFIG_CACULE_SCHED + + #ifdef CONFIG_SCHED_DEBUG + unsigned int nr_spread_over; +diff --git a/kernel/sysctl.c b/kernel/sysctl.c +index d4a78e08f6d8..d8329733e93a 100644 +--- a/kernel/sysctl.c ++++ b/kernel/sysctl.c +@@ -1736,6 +1736,50 @@ static struct ctl_table kern_table[] = { + .mode = 0644, + .proc_handler = proc_dointvec, + }, ++#ifdef CONFIG_CACULE_SCHED ++ { ++ .procname = "sched_interactivity_factor", ++ .data = &interactivity_factor, ++ .maxlen = sizeof(unsigned int), ++ .mode = 0644, ++ .proc_handler = proc_dointvec, ++ }, ++ { ++ .procname = "sched_max_lifetime_ms", ++ .data = &cacule_max_lifetime, ++ .maxlen = sizeof(unsigned int), ++ .mode = 0644, ++ .proc_handler = proc_dointvec, ++ }, ++ { ++ .procname = "sched_cache_factor", ++ .data = &cache_factor, ++ .maxlen = sizeof(unsigned int), ++ .mode = 0644, ++ .proc_handler = proc_dointvec, ++ }, ++ { ++ .procname = "sched_cache_divisor", ++ .data = &cache_divisor, ++ .maxlen = sizeof(unsigned int), ++ .mode = 0644, ++ .proc_handler = proc_dointvec, ++ }, ++ { ++ .procname = "sched_starve_factor", ++ .data = &starve_factor, ++ .maxlen = sizeof(unsigned int), ++ .mode = 0644, ++ .proc_handler = proc_dointvec, ++ }, ++ { ++ .procname = "sched_starve_divisor", ++ .data = &starve_divisor, ++ .maxlen = sizeof(unsigned int), ++ .mode = 0644, ++ .proc_handler = proc_dointvec, ++ }, ++#endif + #ifdef CONFIG_SCHEDSTATS + { + .procname = "sched_schedstats", diff --git a/patches/CacULE/experimental/rdb-5.13-mfs.patch b/patches/CacULE/experimental/rdb-5.13-mfs.patch new file mode 100644 index 000000000..9d6452144 --- /dev/null +++ b/patches/CacULE/experimental/rdb-5.13-mfs.patch @@ -0,0 +1,918 @@ +diff --git a/init/Kconfig b/init/Kconfig +index 089e3bfe5dbc..7c46008e5a7b 100644 +--- a/init/Kconfig ++++ b/init/Kconfig +@@ -844,6 +844,40 @@ config CACULE_SCHED + + If unsure, say Y here. + ++config CACULE_RDB ++ bool "RDB (Response Driven Balancer)" ++ default y ++ depends on CACULE_SCHED ++ help ++ This is an experimental load balancer for CacULE. It is a lightweight ++ load balancer which is a replacement of CFS load balancer. It migrates ++ tasks based on their interactivity scores. ++ ++ If unsure, say Y here. ++ ++config RDB_INTERVAL ++ int "RDB load balancer interval" ++ default 19 ++ depends on CACULE_RDB ++ help ++ This is an interval to control load balance time period. ++ The trigger_load_balance runs in every tick. For High HZ values, the ++ load balance could be overwhelming. RDB load balance includes rq locking ++ which can reduce the performance. The balance interval can help to avoid ++ running load balance on every tick. For example, RDB_INTERVAL=3 will ++ only run load balance every 3ms. Setting RDB_INTERVAL depends on HZ. ++ If you want load balancer run every 2ms while HZ=500 then it is not ++ needed and better to set RDB_INTERVAL=0 since 500HZ already (1000ms ++ / 500HZ = 2ms). However, if you have 1000HZ and want to avoid load ++ balancer from running every 1ms, you could set RDB_INTERVAL=4ms for ++ example to make load balancer run every 4ms. Less RDB_INTERVAL values ++ (or 0 to disable) could make sure tasks are balanced ASAP, but with ++ the cost of locking/blocking time. High RDB_INTERVAL values can relax ++ balancing locking but with the cost of imbalanced workload for that ++ period of time (i.e. if RDB_INTERVAL=100ms) there will be no balancing ++ for 100ms (except for newidle_balance which is not effected by RDB_INTERVAL). ++ ++ If in doubt, use the default value. + + # + # For architectures that want to enable the support for NUMA-affine scheduler +@@ -962,6 +996,7 @@ config CGROUP_WRITEBACK + + menuconfig CGROUP_SCHED + bool "CPU controller" ++ depends on !CACULE_RDB + default n + help + This feature lets CPU scheduler recognize task groups and control CPU +@@ -1239,10 +1274,10 @@ config CHECKPOINT_RESTORE + + config SCHED_AUTOGROUP + bool "Automatic process group scheduling" ++ depends on !CACULE_RDB + select CGROUPS + select CGROUP_SCHED + select FAIR_GROUP_SCHED +- default y + help + This option optimizes the scheduler for common desktop workloads by + automatically creating and populating task groups. This separation +diff --git a/kernel/sched/core.c b/kernel/sched/core.c +index bdedde199504..18d2b5d41b36 100644 +--- a/kernel/sched/core.c ++++ b/kernel/sched/core.c +@@ -8129,8 +8129,11 @@ void __init sched_init(void) + BUG_ON(&dl_sched_class + 1 != &stop_sched_class); + #endif + +-#ifdef CONFIG_CACULE_SCHED +- printk(KERN_INFO "CacULE CPU scheduler v5.13-r2 by Hamad Al Marri."); ++#ifdef CONFIG_CACULE_RDB ++ /* ++ * This patch is on top cacule-5.13.patch ++ */ ++ printk(KERN_INFO "CacULE CPU scheduler (RDB) v5.13-r2 by Hamad Al Marri."); + #endif + + wait_bit_init(); +diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c +index c64337fd121d..e64aa617b3ed 100644 +--- a/kernel/sched/fair.c ++++ b/kernel/sched/fair.c +@@ -841,6 +841,10 @@ static void __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *_se) + cfs_rq->head = NULL; + cfs_rq->tail = NULL; + ++#ifdef CONFIG_CACULE_RDB ++ WRITE_ONCE(cfs_rq->IS_head, ~0); ++#endif ++ + } else if (se == cfs_rq->head) { + // if it is the head + cfs_rq->head = cfs_rq->head->next; +@@ -4562,7 +4566,9 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial) + } + #endif /* CONFIG_CACULE_SCHED */ + ++#if !defined(CONFIG_CACULE_RDB) + static void check_enqueue_throttle(struct cfs_rq *cfs_rq); ++#endif + + static inline void check_schedstat_required(void) + { +@@ -4671,6 +4677,7 @@ enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) + __enqueue_entity(cfs_rq, se); + se->on_rq = 1; + ++#if !defined(CONFIG_CACULE_RDB) + /* + * When bandwidth control is enabled, cfs might have been removed + * because of a parent been throttled but cfs->nr_running > 1. Try to +@@ -4681,6 +4688,7 @@ enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) + + if (cfs_rq->nr_running == 1) + check_enqueue_throttle(cfs_rq); ++#endif + } + + #if !defined(CONFIG_CACULE_SCHED) +@@ -5223,6 +5231,9 @@ static int tg_throttle_down(struct task_group *tg, void *data) + + static bool throttle_cfs_rq(struct cfs_rq *cfs_rq) + { ++#ifdef CONFIG_CACULE_RDB ++ return false; ++#else + struct rq *rq = rq_of(cfs_rq); + struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg); + struct sched_entity *se; +@@ -5300,10 +5311,12 @@ static bool throttle_cfs_rq(struct cfs_rq *cfs_rq) + cfs_rq->throttled = 1; + cfs_rq->throttled_clock = rq_clock(rq); + return true; ++#endif + } + + void unthrottle_cfs_rq(struct cfs_rq *cfs_rq) + { ++#if !defined(CONFIG_CACULE_RDB) + struct rq *rq = rq_of(cfs_rq); + struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg); + struct sched_entity *se; +@@ -5385,6 +5398,7 @@ void unthrottle_cfs_rq(struct cfs_rq *cfs_rq) + /* Determine whether we need to wake up potentially idle CPU: */ + if (rq->curr == rq->idle && rq->cfs.nr_running) + resched_curr(rq); ++#endif + } + + static void distribute_cfs_runtime(struct cfs_bandwidth *cfs_b) +@@ -5837,7 +5851,11 @@ static inline bool cfs_bandwidth_used(void) + + static void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec) {} + static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq) { return false; } ++ ++#if !defined(CONFIG_CACULE_RDB) + static void check_enqueue_throttle(struct cfs_rq *cfs_rq) {} ++#endif ++ + static inline void sync_throttle(struct task_group *tg, int cpu) {} + static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq) {} + +@@ -6081,7 +6099,9 @@ static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags) + struct sched_entity *se = &p->se; + int task_sleep = flags & DEQUEUE_SLEEP; + int idle_h_nr_running = task_has_idle_policy(p); ++#if !defined(CONFIG_CACULE_RDB) + bool was_sched_idle = sched_idle_rq(rq); ++#endif + + util_est_dequeue(&rq->cfs, p); + +@@ -6132,9 +6152,11 @@ static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags) + /* At this point se is NULL and we are at root level*/ + sub_nr_running(rq, 1); + ++#if !defined(CONFIG_CACULE_RDB) + /* balance early to pull high priority tasks */ + if (unlikely(!was_sched_idle && sched_idle_rq(rq))) + rq->next_balance = jiffies; ++#endif + + dequeue_throttle: + util_est_update(&rq->cfs, p, task_sleep); +@@ -7659,11 +7681,23 @@ pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf + if (prev) + put_prev_task(rq, prev); + ++#ifdef CONFIG_CACULE_RDB ++ se = pick_next_entity(cfs_rq, NULL); ++ set_next_entity(cfs_rq, se); ++ ++ if (cfs_rq->head) { ++ unsigned int IS_head = calc_interactivity(sched_clock(), cfs_rq->head); ++ WRITE_ONCE(cfs_rq->IS_head, IS_head); ++ } else { ++ WRITE_ONCE(cfs_rq->IS_head, ~0); ++ } ++#else + do { + se = pick_next_entity(cfs_rq, NULL); + set_next_entity(cfs_rq, se); + cfs_rq = group_cfs_rq(se); + } while (cfs_rq); ++#endif + + p = task_of(se); + +@@ -7685,6 +7719,10 @@ done: __maybe_unused; + return p; + + idle: ++#ifdef CONFIG_CACULE_RDB ++ WRITE_ONCE(cfs_rq->IS_head, ~0); ++#endif ++ + if (!rf) + return NULL; + +@@ -7991,6 +8029,7 @@ struct lb_env { + struct list_head tasks; + }; + ++#if !defined(CONFIG_CACULE_RDB) + /* + * Is this task likely cache-hot: + */ +@@ -8412,6 +8451,7 @@ static void attach_tasks(struct lb_env *env) + + rq_unlock(env->dst_rq, &rf); + } ++#endif + + #ifdef CONFIG_NO_HZ_COMMON + static inline bool cfs_rq_has_blocked(struct cfs_rq *cfs_rq) +@@ -8461,6 +8501,7 @@ static inline void update_blocked_load_tick(struct rq *rq) {} + static inline void update_blocked_load_status(struct rq *rq, bool has_blocked) {} + #endif + ++#if !defined(CONFIG_CACULE_RDB) + static bool __update_blocked_others(struct rq *rq, bool *done) + { + const struct sched_class *curr_class; +@@ -8486,6 +8527,7 @@ static bool __update_blocked_others(struct rq *rq, bool *done) + + return decayed; + } ++#endif + + #ifdef CONFIG_FAIR_GROUP_SCHED + +@@ -8576,6 +8618,7 @@ static unsigned long task_h_load(struct task_struct *p) + cfs_rq_load_avg(cfs_rq) + 1); + } + #else ++#if !defined(CONFIG_CACULE_RDB) + static bool __update_blocked_fair(struct rq *rq, bool *done) + { + struct cfs_rq *cfs_rq = &rq->cfs; +@@ -8587,6 +8630,7 @@ static bool __update_blocked_fair(struct rq *rq, bool *done) + + return decayed; + } ++#endif + + static unsigned long task_h_load(struct task_struct *p) + { +@@ -8594,6 +8638,7 @@ static unsigned long task_h_load(struct task_struct *p) + } + #endif + ++#if !defined(CONFIG_CACULE_RDB) + static void update_blocked_averages(int cpu) + { + bool decayed = false, done = true; +@@ -8612,6 +8657,7 @@ static void update_blocked_averages(int cpu) + cpufreq_update_util(rq, 0); + rq_unlock_irqrestore(rq, &rf); + } ++#endif + + /********** Helpers for find_busiest_group ************************/ + +@@ -9715,6 +9761,7 @@ static inline void calculate_imbalance(struct lb_env *env, struct sd_lb_stats *s + * different in groups. + */ + ++#if !defined(CONFIG_CACULE_RDB) + /** + * find_busiest_group - Returns the busiest group within the sched_domain + * if there is an imbalance. +@@ -9983,6 +10030,7 @@ static struct rq *find_busiest_queue(struct lb_env *env, + + return busiest; + } ++#endif + + /* + * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but +@@ -10019,6 +10067,7 @@ imbalanced_active_balance(struct lb_env *env) + return 0; + } + ++#if !defined(CONFIG_CACULE_RDB) + static int need_active_balance(struct lb_env *env) + { + struct sched_domain *sd = env->sd; +@@ -10351,6 +10400,7 @@ static int load_balance(int this_cpu, struct rq *this_rq, + out: + return ld_moved; + } ++#endif + + static inline unsigned long + get_sd_balance_interval(struct sched_domain *sd, int cpu_busy) +@@ -10389,6 +10439,7 @@ update_next_balance(struct sched_domain *sd, unsigned long *next_balance) + *next_balance = next; + } + ++#if !defined(CONFIG_CACULE_RDB) + /* + * active_load_balance_cpu_stop is run by the CPU stopper. It pushes + * running tasks off the busiest CPU onto idle CPUs. It requires at +@@ -10474,6 +10525,7 @@ static int active_load_balance_cpu_stop(void *data) + } + + static DEFINE_SPINLOCK(balancing); ++#endif + + /* + * Scale the max load_balance interval with the number of CPUs in the system. +@@ -10484,6 +10536,7 @@ void update_max_interval(void) + max_load_balance_interval = HZ*num_online_cpus()/10; + } + ++#if !defined(CONFIG_CACULE_RDB) + /* + * It checks each scheduling domain to see if it is due to be balanced, + * and initiates a balancing operation if so. +@@ -10576,6 +10629,7 @@ static void rebalance_domains(struct rq *rq, enum cpu_idle_type idle) + rq->next_balance = next_balance; + + } ++#endif + + static inline int on_null_domain(struct rq *rq) + { +@@ -10609,6 +10663,7 @@ static inline int find_new_ilb(void) + return nr_cpu_ids; + } + ++#if !defined(CONFIG_CACULE_RDB) + /* + * Kick a CPU to do the nohz balancing, if it is time for it. We pick any + * idle CPU in the HK_FLAG_MISC housekeeping set (if there is one). +@@ -10759,6 +10814,7 @@ static void nohz_balancer_kick(struct rq *rq) + if (flags) + kick_ilb(flags); + } ++#endif /* CONFIG_CACULE_RDB */ + + static void set_cpu_sd_state_busy(int cpu) + { +@@ -10879,11 +10935,17 @@ static bool update_nohz_stats(struct rq *rq) + if (!time_after(jiffies, READ_ONCE(rq->last_blocked_load_update_tick))) + return true; + ++#if !defined(CONFIG_CACULE_RDB) + update_blocked_averages(cpu); ++#endif + + return rq->has_blocked_load; + } + ++#ifdef CONFIG_CACULE_RDB ++static int idle_try_pull_any(struct cfs_rq *cfs_rq); ++#endif ++ + /* + * Internal function that runs load balance for all idle cpus. The load balance + * can be a simple update of blocked load or a complete load balance with +@@ -10953,7 +11015,11 @@ static void _nohz_idle_balance(struct rq *this_rq, unsigned int flags, + rq_unlock_irqrestore(rq, &rf); + + if (flags & NOHZ_BALANCE_KICK) ++#if !defined(CONFIG_CACULE_RDB) + rebalance_domains(rq, CPU_IDLE); ++#else ++ idle_try_pull_any(&rq->cfs); ++#endif + } + + if (time_after(next_balance, rq->next_balance)) { +@@ -10979,6 +11045,7 @@ static void _nohz_idle_balance(struct rq *this_rq, unsigned int flags, + WRITE_ONCE(nohz.has_blocked, 1); + } + ++#if !defined(CONFIG_CACULE_RDB) + /* + * In CONFIG_NO_HZ_COMMON case, the idle balance kickee will do the + * rebalancing for all the cpus for whom scheduler ticks are stopped. +@@ -10999,6 +11066,7 @@ static bool nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle) + + return true; + } ++#endif + + /* + * Check if we need to run the ILB for updating blocked load before entering +@@ -11048,6 +11116,7 @@ static void nohz_newidle_balance(struct rq *this_rq) + } + + #else /* !CONFIG_NO_HZ_COMMON */ ++#if !defined(CONFIG_CACULE_RDB) + static inline void nohz_balancer_kick(struct rq *rq) { } + + static inline bool nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle) +@@ -11056,8 +11125,130 @@ static inline bool nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle + } + + static inline void nohz_newidle_balance(struct rq *this_rq) { } ++#endif ++ + #endif /* CONFIG_NO_HZ_COMMON */ + ++#ifdef CONFIG_CACULE_RDB ++static int ++can_migrate_task(struct task_struct *p, int dst_cpu, struct rq *src_rq) ++{ ++ if (task_running(src_rq, p)) ++ return 0; ++ ++ if (!cpumask_test_cpu(dst_cpu, p->cpus_ptr)) ++ return 0; ++ ++ if (p->se.exec_start == 0) ++ return 0; ++ ++ return 1; ++} ++ ++static void push_to_unlock(struct rq *this_rq, ++ struct rq *dst_rq, ++ struct task_struct *p, ++ int dst_cpu) ++{ ++ struct rq_flags rf; ++ ++ // detach task ++ deactivate_task(this_rq, p, DEQUEUE_NOCLOCK); ++ set_task_cpu(p, dst_cpu); ++ ++ // unlock this rq ++ raw_spin_unlock(&this_rq->lock); ++ ++ /* push to */ ++ rq_lock_irqsave(dst_rq, &rf); ++ update_rq_clock(dst_rq); ++ ++ activate_task(dst_rq, p, ENQUEUE_NOCLOCK); ++ check_preempt_curr(dst_rq, p, 0); ++ ++ // unlock src rq ++ rq_unlock(dst_rq, &rf); ++ local_irq_restore(rf.flags); ++} ++ ++static void pull_from_unlock(struct rq *this_rq, ++ struct rq *src_rq, ++ struct rq_flags *rf, ++ struct task_struct *p, ++ int dst_cpu) ++{ ++ // detach task ++ deactivate_task(src_rq, p, DEQUEUE_NOCLOCK); ++ set_task_cpu(p, dst_cpu); ++ ++ // unlock src rq ++ rq_unlock(src_rq, rf); ++ local_irq_restore(rf->flags); ++ ++ // lock this rq ++ raw_spin_lock(&this_rq->lock); ++ update_rq_clock(this_rq); ++ ++ activate_task(this_rq, p, ENQUEUE_NOCLOCK); ++ check_preempt_curr(this_rq, p, 0); ++ ++ // unlock this rq ++ raw_spin_unlock(&this_rq->lock); ++} ++ ++static inline struct rq * ++find_max_IS_rq(struct cfs_rq *cfs_rq, int dst_cpu) ++{ ++ struct rq *tmp_rq, *max_rq = NULL; ++ int cpu; ++ unsigned int max_IS = cfs_rq->IS_head; ++ unsigned int local_IS; ++ ++ // find max hrrn ++ for_each_online_cpu(cpu) { ++ if (cpu == dst_cpu) ++ continue; ++ ++ tmp_rq = cpu_rq(cpu); ++ ++ if (tmp_rq->cfs.nr_running < 2 || !tmp_rq->cfs.head) ++ continue; ++ ++ local_IS = READ_ONCE(tmp_rq->cfs.IS_head); ++ ++ if (local_IS < max_IS) { ++ max_IS = local_IS; ++ max_rq = tmp_rq; ++ } ++ } ++ ++ return max_rq; ++} ++ ++static int try_pull_from(struct rq *src_rq, struct rq *this_rq) ++{ ++ struct rq_flags rf; ++ int dst_cpu = cpu_of(this_rq); ++ struct task_struct *p; ++ ++ rq_lock_irqsave(src_rq, &rf); ++ update_rq_clock(src_rq); ++ ++ if (src_rq->cfs.head && src_rq->cfs.nr_running > 1) { ++ p = task_of(se_of(src_rq->cfs.head)); ++ ++ if (can_migrate_task(p, dst_cpu, src_rq)) { ++ pull_from_unlock(this_rq, src_rq, &rf, p, dst_cpu); ++ return 1; ++ } ++ } ++ ++ rq_unlock(src_rq, &rf); ++ local_irq_restore(rf.flags); ++ ++ return 0; ++} ++ + /* + * newidle_balance is called by schedule() if this_cpu is about to become + * idle. Attempts to pull tasks from other CPUs. +@@ -11068,6 +11259,107 @@ static inline void nohz_newidle_balance(struct rq *this_rq) { } + * > 0 - success, new (fair) tasks present + */ + static int newidle_balance(struct rq *this_rq, struct rq_flags *rf) ++{ ++ int this_cpu = this_rq->cpu; ++ struct task_struct *p = NULL; ++ struct rq *src_rq; ++ int src_cpu; ++ struct rq_flags src_rf; ++ int pulled_task = 0; ++ int cores_round = 1; ++ ++ update_misfit_status(NULL, this_rq); ++ /* ++ * We must set idle_stamp _before_ calling idle_balance(), such that we ++ * measure the duration of idle_balance() as idle time. ++ */ ++ this_rq->idle_stamp = rq_clock(this_rq); ++ ++ /* ++ * Do not pull tasks towards !active CPUs... ++ */ ++ if (!cpu_active(this_cpu)) ++ return 0; ++ ++ /* ++ * This is OK, because current is on_cpu, which avoids it being picked ++ * for load-balance and preemption/IRQs are still disabled avoiding ++ * further scheduler activity on it and we're being very careful to ++ * re-start the picking loop. ++ */ ++ rq_unpin_lock(this_rq, rf); ++ raw_spin_unlock(&this_rq->lock); ++ ++again: ++ for_each_online_cpu(src_cpu) { ++ ++ if (src_cpu == this_cpu) ++ continue; ++ ++ if (cores_round && !cpus_share_cache(src_cpu, this_cpu)) ++ continue; ++ ++ src_rq = cpu_rq(src_cpu); ++ ++ rq_lock_irqsave(src_rq, &src_rf); ++ update_rq_clock(src_rq); ++ ++ if (src_rq->cfs.nr_running < 2 || !(src_rq->cfs.head)) ++ goto next; ++ ++ p = task_of(se_of(src_rq->cfs.head)); ++ ++ if (can_migrate_task(p, this_cpu, src_rq)) { ++ pull_from_unlock(this_rq, src_rq, &src_rf, p, this_cpu); ++ ++ pulled_task = 1; ++ goto out; ++ } ++ ++next: ++ rq_unlock(src_rq, &src_rf); ++ local_irq_restore(src_rf.flags); ++ ++ /* ++ * Stop searching for tasks to pull if there are ++ * now runnable tasks on this rq. ++ */ ++ if (pulled_task || this_rq->nr_running > 0) ++ goto out; ++ } ++ ++ if (cores_round) { ++ // now search for all cpus ++ cores_round = 0; ++ goto again; ++ } ++ ++out: ++ raw_spin_lock(&this_rq->lock); ++ ++ /* ++ * While browsing the domains, we released the rq lock, a task could ++ * have been enqueued in the meantime. Since we're not going idle, ++ * pretend we pulled a task. ++ */ ++ if (this_rq->cfs.h_nr_running && !pulled_task) ++ pulled_task = 1; ++ ++ /* Is there a task of a high priority class? */ ++ if (this_rq->nr_running != this_rq->cfs.h_nr_running) ++ pulled_task = -1; ++ ++ if (pulled_task) ++ this_rq->idle_stamp = 0; ++ else ++ nohz_newidle_balance(this_rq); ++ ++ rq_repin_lock(this_rq, rf); ++ ++ return pulled_task; ++} ++#else ++static int newidle_balance(struct rq *this_rq, struct rq_flags *rf) + { + unsigned long next_balance = jiffies + HZ; + int this_cpu = this_rq->cpu; +@@ -11224,6 +11516,214 @@ void trigger_load_balance(struct rq *rq) + + nohz_balancer_kick(rq); + } ++#endif ++ ++#ifdef CONFIG_CACULE_RDB ++static int ++idle_try_pull_any(struct cfs_rq *cfs_rq) ++{ ++ struct task_struct *p = NULL; ++ struct rq *this_rq = rq_of(cfs_rq), *src_rq; ++ int dst_cpu = cpu_of(this_rq); ++ int src_cpu; ++ struct rq_flags rf; ++ int pulled = 0; ++ int cores_round = 1; ++ ++again: ++ for_each_online_cpu(src_cpu) { ++ ++ if (src_cpu == dst_cpu) ++ continue; ++ ++ if (cores_round && !cpus_share_cache(src_cpu, dst_cpu)) ++ continue; ++ ++ src_rq = cpu_rq(src_cpu); ++ ++ rq_lock_irqsave(src_rq, &rf); ++ update_rq_clock(src_rq); ++ ++ if (src_rq->cfs.nr_running < 2 || !(src_rq->cfs.head)) ++ goto next; ++ ++ p = task_of(se_of(src_rq->cfs.head)); ++ ++ if (can_migrate_task(p, dst_cpu, src_rq)) { ++ pull_from_unlock(this_rq, src_rq, &rf, p, dst_cpu); ++ pulled = 1; ++ goto out; ++ } ++ ++next: ++ rq_unlock(src_rq, &rf); ++ local_irq_restore(rf.flags); ++ } ++ ++ if (cores_round) { ++ // now search for all cpus ++ cores_round = 0; ++ goto again; ++ } ++ ++out: ++ return pulled; ++} ++ ++ ++static int ++try_pull_higher_IS(struct cfs_rq *cfs_rq) ++{ ++ struct rq *this_rq = rq_of(cfs_rq), *max_rq; ++ int dst_cpu = cpu_of(this_rq); ++ ++ max_rq = find_max_IS_rq(cfs_rq, dst_cpu); ++ ++ if (!max_rq) ++ return 0; ++ ++ if (try_pull_from(max_rq, this_rq)) ++ return 1; ++ ++ return 0; ++} ++ ++static void try_push_any(struct rq *this_rq) ++{ ++ struct task_struct *p = NULL; ++ struct rq *dst_rq; ++ int dst_cpu; ++ int src_cpu = cpu_of(this_rq); ++ int cores_round = 1; ++ ++again: ++ for_each_online_cpu(dst_cpu) { ++ ++ if (dst_cpu == src_cpu) ++ continue; ++ ++ if (cores_round && !cpus_share_cache(src_cpu, dst_cpu)) ++ continue; ++ ++ dst_rq = cpu_rq(dst_cpu); ++ ++ if (dst_rq->cfs.nr_running >= this_rq->cfs.nr_running - 1) ++ continue; ++ ++ // lock this rq ++ raw_spin_lock(&this_rq->lock); ++ update_rq_clock(this_rq); ++ ++ if (!this_rq->cfs.head) { ++ // unlock this rq ++ raw_spin_unlock(&this_rq->lock); ++ return; ++ } ++ ++ p = task_of(se_of(this_rq->cfs.head)); ++ ++ if (can_migrate_task(p, dst_cpu, this_rq)) { ++ push_to_unlock(this_rq, dst_rq, p, dst_cpu); ++ return; ++ } ++ ++ // unlock this rq ++ raw_spin_unlock(&this_rq->lock); ++ } ++ ++ if (cores_round) { ++ // now search for all cpus ++ cores_round = 0; ++ goto again; ++ } ++} ++ ++static void try_pull_any(struct rq *this_rq) ++{ ++ struct task_struct *p = NULL; ++ struct rq *src_rq; ++ int dst_cpu = cpu_of(this_rq); ++ int src_cpu; ++ struct rq_flags src_rf; ++ int cores_round = 1; ++ unsigned int this_head = this_rq->cfs.IS_head; ++ ++again: ++ for_each_online_cpu(src_cpu) { ++ ++ if (src_cpu == dst_cpu) ++ continue; ++ ++ if (cores_round && !cpus_share_cache(src_cpu, dst_cpu)) ++ continue; ++ ++ src_rq = cpu_rq(src_cpu); ++ ++ if (src_rq->cfs.nr_running < 2 || !(src_rq->cfs.head) ++ || READ_ONCE(src_rq->cfs.IS_head) >= this_head) ++ continue; ++ ++ rq_lock_irqsave(src_rq, &src_rf); ++ update_rq_clock(src_rq); ++ ++ if (src_rq->cfs.nr_running < 2 || !(src_rq->cfs.head) ++ || src_rq->cfs.IS_head >= this_head) ++ goto next; ++ ++ p = task_of(se_of(src_rq->cfs.head)); ++ ++ if (can_migrate_task(p, dst_cpu, src_rq)) { ++ pull_from_unlock(this_rq, src_rq, &src_rf, p, dst_cpu); ++ return; ++ } ++ ++next: ++ rq_unlock(src_rq, &src_rf); ++ local_irq_restore(src_rf.flags); ++ } ++ ++ if (cores_round) { ++ // now search for all cpus ++ cores_round = 0; ++ goto again; ++ } ++} ++ ++static inline void ++active_balance(struct rq *rq) ++{ ++ struct cfs_rq *cfs_rq = &rq->cfs; ++ ++ if (!cfs_rq->head || cfs_rq->nr_running < 2) ++ try_pull_higher_IS(&rq->cfs); ++ else { ++ try_push_any(rq); ++ try_pull_any(rq); ++ } ++} ++ ++void trigger_load_balance(struct rq *rq) ++{ ++ unsigned long interval; ++ ++#ifdef CONFIG_RDB_INTERVAL ++ if (time_before(jiffies, rq->next_balance)) ++ return; ++#endif ++ ++ if (rq->idle_balance) ++ idle_try_pull_any(&rq->cfs); ++ else { ++ active_balance(rq); ++ ++#ifdef CONFIG_RDB_INTERVAL ++ /* scale ms to jiffies */ ++ interval = msecs_to_jiffies(CONFIG_RDB_INTERVAL); ++ rq->next_balance = jiffies + interval; ++#endif ++ } ++} ++#endif + + static void rq_online_fair(struct rq *rq) + { +@@ -11864,7 +12364,9 @@ void show_numa_stats(struct task_struct *p, struct seq_file *m) + __init void init_sched_fair_class(void) + { + #ifdef CONFIG_SMP ++#if !defined(CONFIG_CACULE_RDB) + open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); ++#endif + + #ifdef CONFIG_NO_HZ_COMMON + nohz.next_balance = jiffies; +diff --git a/kernel/sched/idle.c b/kernel/sched/idle.c +index 7ca3d3d86c2a..a7422dea8a9f 100644 +--- a/kernel/sched/idle.c ++++ b/kernel/sched/idle.c +@@ -262,10 +262,12 @@ static void do_idle(void) + { + int cpu = smp_processor_id(); + ++#if !defined(CONFIG_CACULE_RDB) + /* + * Check if we need to update blocked load + */ + nohz_run_idle_balance(cpu); ++#endif + + /* + * If the arch has a polling bit, we maintain an invariant: +diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h +index 09a8290fc883..4d46ef7190bf 100644 +--- a/kernel/sched/sched.h ++++ b/kernel/sched/sched.h +@@ -544,6 +544,10 @@ struct cfs_rq { + struct cacule_node *head; + struct cacule_node *tail; + ++#ifdef CONFIG_CACULE_RDB ++ unsigned int IS_head; ++#endif ++ + #else + struct sched_entity *next; + struct sched_entity *last;